基于双模态磁共振成像和决策层融合的抑郁症辅助诊断

本研究提出一种基于结构和功能双模态磁共振成像数据融合的抑郁症分类算法，首先利用功能脑网络和深度学习网络分别提取功能和结构磁共振成像数据特征，并计算类概率，然后使用软投票法和加权投票法在决策层对两种类概率数据进行融合，充分提取功能与结构磁共振成像的数据信息，得到更加准确的分类效果。试验结果表明，数据融合方法可以显著提高抑郁症分类效果，获得91.34%的准确率和96.62%的召回率，更好地实现了抑郁症的辅助诊断与预后。

Auxiliary diagnosis of depression based on bimodal magnetic resonance imaging and decision level fusion

Abstract: A depression classification algorithm based on structural and functional magnetic resonance imaging data fusion is proposed. After extracting functional and structural data features by functional brain network and deep learning network, and obtaining class probability, soft voting method and weighted voting method are used to fuse two kinds of probability data at decision level for fully extracting the data information of functional and structural magnetic resonance imaging, thereby obtaining more accurate classification results. The test results show that the data fusion method can effectively improve the performance in depression classification, achieving an accuracy of 91.34% and a recall rate of 96.62%, which testifies that the proposed method can better realize the auxiliary diagnosis and prognosis of depression.